The future of AI is not just software. It is robots operating in the real world. Sunny Cheung examines how China's push to deploy physical AI at scale is reshaping global competition and why the United States must strengthen its robotics ecosystem to keep pace.
In June 2026, China’s Ministry of Industry and Information Technology and the State-owned Assets Supervision and Administration Commission launched a nationwide initiative to move humanoid robots and embodied artificial intelligence out of laboratories and demonstration halls and into factories, warehouses, hospitals, stores, and other real workplaces.
China’s robotics targets are ambitious. By the end of 2026, Beijing plans to identify more than 100 high-value applications and build the capacity to deploy humanoid robots at the scale of tens of thousands of units. Provincial governments and centrally controlled state-owned enterprises must construct real operating environments, organize partnerships among robot makers and users, and evaluate whether the machines improve productivity, safety, reliability, and economic performance.
The significance is not simply that China is subsidizing another emerging industry. Beijing is now mobilizing anchor customers—especially state-owned enterprises—to create demand, generate operational data, and help robotics firms cross the difficult gap between an impressive prototype and a commercially useful product. The United States must address this gap.
Artificial intelligence has so far been associated primarily with software. Physical AI combines AI models with sensors, actuators, mobility systems, machine vision, and control software, enabling machines to perceive their surroundings and act within them.
A robot must learn how objects move, how much force to apply, how to recover from errors, and how to operate safely around people and equipment. That requires physical-world data generated through repeated deployment.
China’s initiative is designed to overcome this challenge. Governments and state-owned enterprises provide operating environments; manufacturers adapt systems to specific tasks; and model developers, suppliers, and researchers support iteration. Successful solutions are then reproduced across similar facilities—a “validate one, deploy many” model.
This creates a powerful flywheel: deployment produces data; data improves models and hardware; better performance lowers adoption risk; and lower risk encourages further deployment.
China has already been building the rest of this ecosystem. A Reuters review found that authorities had directed more than $20 billion toward humanoid robotics and related technologies, while government procurement of humanoid robots and associated systems rose from approximately 4.7 million yuan in 2023 to 214 million yuan in 2024. Municipalities have also offered funds, facilities, workspace, and data-center support.
In January 2026, eight Chinese departments issued an “AI Plus Manufacturing” action plan calling for pilot-production facilities, training grounds, benchmark lines, and early manufacturing deployments. A Beijing data and training base reported more than 120 robots across over 30 real-world scenarios, ranging from industrial production and retail to medicine and eldercare.
Commercial momentum is also visible. In March, Unitree filed for a Shanghai Stock Exchange listing after reporting more than 5,500 humanoid shipments in 2025 in its IPO disclosures. On June 18, the Ministry of Commerce also announced new measures to promote AI-enabled consumption, including the development of a market for humanoid robots. The June deployment initiative now connects these investments to users with factories, infrastructure, and purchasing power. The state is no longer merely funding supply; it is organizing demand.
China’s interest reflects both necessity and ambition. Its population is aging, its workforce is contracting, and domestic manufacturers face rising labor costs and intensifying competition from lower-cost producers in Southeast Asia and India. Robots handling repetitive, dangerous, and/or precision work could help preserve China’s manufacturing competitiveness.
Physical AI could also augment China’s existing industrial advantages. Dense networks of electronics, machinery, battery, logistics, and component suppliers can lower costs and accelerate iteration. Leadership in actuators, sensors, hands, motors, controls, and software would reinforce China’s position across advanced manufacturing.
The implications extend beyond humanoids to vehicles, drones, warehouses, agriculture, construction, medicine, and defense. AI is expected to reshape productivity and work across the global economy; the International Monetary Fund estimates that it could affect nearly 40 percent of jobs worldwide. Physical AI brings that transformation onto factory floors, logistics networks, infrastructure, and potentially even battlefields.
For the United States, this is not a contest over robots walking across a stage. It is a competition over the productivity, resilience, and technological depth of the industrial economy.
The United States retains strengths in frontier models, semiconductors, robotics research, software, capital, and engineering. American companies are developing capable robots for logistics, manufacturing, defense, healthcare, and agriculture.
Yet companies are often stranded between research and scale. Robots remain expensive, manufacturers resist reorganizing facilities around immature machines, and startups cannot gather enough operational data without sustained deployments. Small pilots rarely provide the order volume or revenue certainty needed for a domestic ecosystem.
Washington has recognized similar problems elsewhere. A June 2025 executive order on American drone dominance directed agencies to accelerate testing, integrate drones into federal operations, and support domestic production. A subsequent government-wide procurement memorandum established security requirements while explicitly supporting US producers.
The Department of War has taken an even more direct approach in critical minerals. Its 2025 public-private agreement with MP Materials combined investment, financing, a ten-year price floor, and a long-term magnet purchase commitment. Government acted as an anchor customer willing to reduce market uncertainty. Robotics deserves the same strategic attention.
First, the White House should establish a national physical AI and robotics strategy spanning the relevant economic, security, infrastructure, health, and labor agencies. It should identify priority applications, supply-chain vulnerabilities, safety requirements, workforce implications, and measurable deployment goals.
Second, federal agencies should become early customers, procuring and testing robots for hazardous-material handling, infrastructure inspection, disaster response, logistics, laboratory automation, nuclear maintenance, wildfire management, and other dangerous or labor-constrained work. Procurement should target operational problems, not a particular humanoid form.
Third, government should establish shared real-world testing environments. National laboratories, depots, ports, hospitals, factories, and infrastructure sites could host controlled deployments that let qualified companies gather data and validate performance, with privacy, cybersecurity, safety, and intellectual-property protections built in.
Fourth, Washington should use milestone-based contracts, advance purchase commitments, and challenge competitions to reward improvements in reliability and cost. Support should depend on measurable performance—task completion, uptime, safety, productivity, and domestic value creation—not theatrical demonstrations.
Finally, the United States should build a secure allied robotics supply chain. Motors, reducers, actuators, sensors, hands, power electronics, batteries, and control hardware deserve the scrutiny now applied to chips, drones, and rare earths.
China is compressing years of market formation into an organized deployment campaign. Projects will fail and many robots will initially perform narrow tasks. But even imperfect deployments will generate experience, data, suppliers, engineers, and customers.
The United States should not imitate China’s economic system, but it should learn from the problem Beijing is solving. Leadership in physical AI will belong not to the country with the best prototype, but to the one that deploys useful machines, learns from them, and builds the surrounding ecosystem the fastest.
The views expressed are the author's alone, and do not represent the views of the Wilson Center.
